iSixSigma

Deviance Test?

Six Sigma – iSixSigma Forums Old Forums General Deviance Test?

Viewing 9 posts - 1 through 9 (of 9 total)
  • Author
    Posts
  • #48954

    Ryan
    Member

    Hello,
    Sorry for the multiple posts…
    I was recently reading an article that stated that they ran a deviance test to test “cohort effects.” They were concerned because the treatment was in a group setting, so each group was labeled a cohort. The “treatment” variable had two levels (intervention and placebo, both of which were administered in group settings). I believe they used a Generalized Linear Modeling approach (commonly referred to as GENMOD in SAS) to test if the treatment was effective on a continous variable over time.
    I’ve never heard of the term “deviance test” and after googling, searching message boards, etc. I found a term called “Analysis of Deviance” or ANODEV. According to what I read ANODEV measures whether a variable is significantly predicting the DV above and beyond the other variables in the model. You compare a full model to a reduced model. I just do not see how that would answer whether cohort effect (a categorical variable) on some variable of interest is significant. Would you not run a nested model where groups are nested within treatments?
    Perhaps there is another deviance test I am not familiar with, or I am simply not making the connection.
    Any thoughts would be greatly appreciated.
    Ryan

    0
    #166334

    Mikel
    Member

    I think the deviance test was created by Darth for his activities on
    SOBE. You’ll have to contact him for details.

    0
    #166339

    Haugen
    Participant

    I believe the Darth nested model where groups are nested within treatments is called the Thong model.

    0
    #166342

    Stevo
    Member

    Darth a thong model, hardly!  Maybe Jocky.
    Stevo

    0
    #166343

    Mikel
    Member

    You must not know the Darth we know (and love).

    0
    #166348

    Brandon
    Participant

    I guarantee you there are a bunch of us in this forum who refuse to take a Deviance Test. Especially since the results may end up in the hands of the authorities.

    0
    #166349

    Robert Butler
    Participant

    From Categorical Data Analysis 2nd Edition -Agresti – pp.118-119 there is a good discussion of deviance and testing.  Since it is long I’ll just note the one sentence that would seem to pertain to your question.
    “Deviance is a test of model fit….[deviance is used] for model checking and for interential comparisons of models. Components of the deviance are residual measures of lack of fit.” 
      So it would appear that all the article is saying is they used measures of deviance to test a model where they included cohort variable(s) vs. a model where they didn’t.

    0
    #166350

    Ryan
    Member

    For those who are interested, I’ve come to learn that the deviance test they were most likely referring to was the Likelihood Ratio Test. I don’t think it’s the best approach to test for groups-within-treatments effects, but it is what it is.  

    0
    #166351

    Ryan
    Member

    Yes. I agree, Robert. Thank you again for all your help. -Ryan

    0
Viewing 9 posts - 1 through 9 (of 9 total)

The forum ‘General’ is closed to new topics and replies.